Helicone

Helicone sits between your application and the model providers. Point an OpenAI SDK at its gateway, keep a single Helicone key, and requests can be routed to more than 100 models across the major providers, with automatic fallback to another provider serving the same model when one fails. The company says it adds no markup on model pricing and routes to the cheapest provider for a given model.

Every request that passes through is logged, which is where Helicone started: a dashboard of requests, errors, latency, cost and top models; session traces that lay out multi-step agent runs as a timeline of LLM and tool calls; user and custom-property analytics; and HQL, a query language over your logs. Around that sit prompt management, a playground, datasets, scores and online evaluators, plus caching, rate limits, alerts and webhooks.

The project is open source under Apache 2.0 and can be self-hosted with Docker. The hosted plans start with a free Hobby tier (10,000 requests a month, one seat), then Pro at $79 a month and Team at $799 a month with usage-based pricing on top; Enterprise adds SAML SSO and on-prem deployment. It suits teams running several models in production who want routing and observability in one layer rather than two.